医学影像处理的深度学习可解释性研究进展

随着医学影像数据的迅速增长，传统的影像分析方法给医生带来巨大挑战。利用计算机视觉技术提供自动或半自动辅助诊断，可大大缓解人工阅片压力，提高诊断的准确性，促进医疗流程的标准化建设等。目前，深度学习卷积神经网络在医学影像处理中已取得不俗表现，但深度学习“黑匣子”的不可解释性阻碍了智能医疗诊断的发展。为增强对医学影像数据处理的深度学习可解释性的了解，对近几年相关研究进展进行了综述。首先，综述了深度学习在医学领域的应用现状及面临的问题，对神经网络的可解释性内涵进行了讨论；然后，从现有深度学习可解释性的常见方法出发，重点讨论了医学影像处理的深度学习可解释性研究进展；最后，探讨了医学影像处理的深度学习可解释性的发展趋势。

A review on deep learning interpretability in medical image processing

Medical image data are rapidly accumulating and traditional image analysis methods based on manual approaches has imposed a heavy burden on doctors. Computer vision has played an important role in alleviating the pressure of manual reading,improving the accuracy of diagnosis and promoting the standardization of medical procedures by providing automatic or semi-automatic auxiliary diagnostic methods. At present,deep learning convolutional neural network has achieved outstanding performance in various medical image processing tasks,but the unexplainability of deep learning "black box" has become a major obstacle to further explore the full potentials of intelligent medical diagnosis.This paper summarizes the research progress of deep learning interpretability in medical image processing in recent years.Firstly,we clarify the application status and problems of deep learning in the medical field,and discusses the interpretable connotation of neural networks. Then,we focus on the research progress of deep learning interpretability in medical image data processing starting from the common methods of deep learning interpretability. Finally,the interpretable development trend of medical image processing is discussed.